mm: add FOLL_MLOCK follow_page flag.
[linux-2.6.git] / mm / page_cgroup.c
index 3dd4a90..5bffada 100644 (file)
@@ -9,6 +9,7 @@
 #include <linux/vmalloc.h>
 #include <linux/cgroup.h>
 #include <linux/swapops.h>
+#include <linux/kmemleak.h>
 
 static void __meminit
 __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn)
@@ -47,8 +48,6 @@ static int __init alloc_node_page_cgroup(int nid)
        struct page_cgroup *base, *pc;
        unsigned long table_size;
        unsigned long start_pfn, nr_pages, index;
-       struct page *page;
-       unsigned int order;
 
        start_pfn = NODE_DATA(nid)->node_start_pfn;
        nr_pages = NODE_DATA(nid)->node_spanned_pages;
@@ -57,13 +56,11 @@ static int __init alloc_node_page_cgroup(int nid)
                return 0;
 
        table_size = sizeof(struct page_cgroup) * nr_pages;
-       order = get_order(table_size);
-       page = alloc_pages_node(nid, GFP_NOWAIT | __GFP_ZERO, order);
-       if (!page)
-               page = alloc_pages_node(-1, GFP_NOWAIT | __GFP_ZERO, order);
-       if (!page)
+
+       base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
+                       table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
+       if (!base)
                return -ENOMEM;
-       base = page_address(page);
        for (index = 0; index < nr_pages; index++) {
                pc = base + index;
                __init_page_cgroup(pc, start_pfn + index);
@@ -73,7 +70,7 @@ static int __init alloc_node_page_cgroup(int nid)
        return 0;
 }
 
-void __init page_cgroup_init(void)
+void __init page_cgroup_init_flatmem(void)
 {
 
        int nid, fail;
@@ -87,12 +84,12 @@ void __init page_cgroup_init(void)
                        goto fail;
        }
        printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
-       printk(KERN_INFO "please try cgroup_disable=memory option if you"
-       " don't want\n");
+       printk(KERN_INFO "please try 'cgroup_disable=memory' option if you"
+       " don't want memory cgroups\n");
        return;
 fail:
-       printk(KERN_CRIT "allocation of page_cgroup was failed.\n");
-       printk(KERN_CRIT "please try cgroup_disable=memory boot option\n");
+       printk(KERN_CRIT "allocation of page_cgroup failed.\n");
+       printk(KERN_CRIT "please try 'cgroup_disable=memory' boot option\n");
        panic("Out of memory");
 }
 
@@ -103,6 +100,8 @@ struct page_cgroup *lookup_page_cgroup(struct page *page)
        unsigned long pfn = page_to_pfn(page);
        struct mem_section *section = __pfn_to_section(pfn);
 
+       if (!section->page_cgroup)
+               return NULL;
        return section->page_cgroup + pfn;
 }
 
@@ -117,16 +116,23 @@ static int __init_refok init_section_page_cgroup(unsigned long pfn)
        if (!section->page_cgroup) {
                nid = page_to_nid(pfn_to_page(pfn));
                table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION;
-               if (slab_is_available()) {
+               VM_BUG_ON(!slab_is_available());
+               if (node_state(nid, N_HIGH_MEMORY)) {
                        base = kmalloc_node(table_size,
-                                       GFP_KERNEL | __GFP_NOWARN, nid);
+                               GFP_KERNEL | __GFP_NOWARN, nid);
                        if (!base)
                                base = vmalloc_node(table_size, nid);
                } else {
-                       base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
-                               table_size,
-                               PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
+                       base = kmalloc(table_size, GFP_KERNEL | __GFP_NOWARN);
+                       if (!base)
+                               base = vmalloc(table_size);
                }
+               /*
+                * The value stored in section->page_cgroup is (base - pfn)
+                * and it does not point to the memory block allocated above,
+                * causing kmemleak false positives.
+                */
+               kmemleak_not_leak(base);
        } else {
                /*
                 * We don't have to allocate page_cgroup again, but
@@ -261,14 +267,14 @@ void __init page_cgroup_init(void)
                fail = init_section_page_cgroup(pfn);
        }
        if (fail) {
-               printk(KERN_CRIT "try cgroup_disable=memory boot option\n");
+               printk(KERN_CRIT "try 'cgroup_disable=memory' boot option\n");
                panic("Out of memory");
        } else {
                hotplug_memory_notifier(page_cgroup_callback, 0);
        }
        printk(KERN_INFO "allocated %ld bytes of page_cgroup\n", total_usage);
-       printk(KERN_INFO "please try cgroup_disable=memory option if you don't"
-       " want\n");
+       printk(KERN_INFO "please try 'cgroup_disable=memory' option if you don't"
+       " want memory cgroups\n");
 }
 
 void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
@@ -285,6 +291,7 @@ static DEFINE_MUTEX(swap_cgroup_mutex);
 struct swap_cgroup_ctrl {
        struct page **map;
        unsigned long length;
+       spinlock_t      lock;
 };
 
 struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES];
@@ -318,8 +325,6 @@ static int swap_cgroup_prepare(int type)
        struct swap_cgroup_ctrl *ctrl;
        unsigned long idx, max;
 
-       if (!do_swap_account)
-               return 0;
        ctrl = &swap_cgroup_ctrl[type];
 
        for (idx = 0; idx < ctrl->length; idx++) {
@@ -338,6 +343,43 @@ not_enough_page:
 }
 
 /**
+ * swap_cgroup_cmpxchg - cmpxchg mem_cgroup's id for this swp_entry.
+ * @end: swap entry to be cmpxchged
+ * @old: old id
+ * @new: new id
+ *
+ * Returns old id at success, 0 at failure.
+ * (There is no mem_cgroup useing 0 as its id)
+ */
+unsigned short swap_cgroup_cmpxchg(swp_entry_t ent,
+                                       unsigned short old, unsigned short new)
+{
+       int type = swp_type(ent);
+       unsigned long offset = swp_offset(ent);
+       unsigned long idx = offset / SC_PER_PAGE;
+       unsigned long pos = offset & SC_POS_MASK;
+       struct swap_cgroup_ctrl *ctrl;
+       struct page *mappage;
+       struct swap_cgroup *sc;
+       unsigned long flags;
+       unsigned short retval;
+
+       ctrl = &swap_cgroup_ctrl[type];
+
+       mappage = ctrl->map[idx];
+       sc = page_address(mappage);
+       sc += pos;
+       spin_lock_irqsave(&ctrl->lock, flags);
+       retval = sc->id;
+       if (retval == old)
+               sc->id = new;
+       else
+               retval = 0;
+       spin_unlock_irqrestore(&ctrl->lock, flags);
+       return retval;
+}
+
+/**
  * swap_cgroup_record - record mem_cgroup for this swp_entry.
  * @ent: swap entry to be recorded into
  * @mem: mem_cgroup to be recorded
@@ -355,17 +397,17 @@ unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id)
        struct page *mappage;
        struct swap_cgroup *sc;
        unsigned short old;
-
-       if (!do_swap_account)
-               return 0;
+       unsigned long flags;
 
        ctrl = &swap_cgroup_ctrl[type];
 
        mappage = ctrl->map[idx];
        sc = page_address(mappage);
        sc += pos;
+       spin_lock_irqsave(&ctrl->lock, flags);
        old = sc->id;
        sc->id = id;
+       spin_unlock_irqrestore(&ctrl->lock, flags);
 
        return old;
 }
@@ -387,9 +429,6 @@ unsigned short lookup_swap_cgroup(swp_entry_t ent)
        struct swap_cgroup *sc;
        unsigned short ret;
 
-       if (!do_swap_account)
-               return 0;
-
        ctrl = &swap_cgroup_ctrl[type];
        mappage = ctrl->map[idx];
        sc = page_address(mappage);
@@ -420,6 +459,7 @@ int swap_cgroup_swapon(int type, unsigned long max_pages)
        mutex_lock(&swap_cgroup_mutex);
        ctrl->length = length;
        ctrl->map = array;
+       spin_lock_init(&ctrl->lock);
        if (swap_cgroup_prepare(type)) {
                /* memory shortage */
                ctrl->map = NULL;